Currently during the reshardCollection operation, the concurrent write operations on the collection being resharded can succeed during this phase execution. The distributed lock won't exclude new incoming writes on the collection. Since the change stream returns the events ordered on the clusterTime, there could be some writes which start using the new shard key value before logging the reshardCollection operation in the oplog. This would result in us reporting write events with new shard key value (as documentKey), before reporting the reshardCollection event.